9 Subsequently, we used mixed effects models that included patient socio-demographic variables as fixed effects plus a random effect for practice.
> -wrap-foot> Both models were adjusted for age, gender, deprivation, self-rated health status, presence of a mental health condition, and a random effect for practice.
The annual rate of patients changing practice without change of address was modelled using Poisson regression with a random effect for practice included to account for any remaining variation in disenrollment rates.
Just over 70% of respondents describe themselves as White British, with 5.1%, 2.7% and 0.7% of respondents describing themselves as Indian, Pakistani and Bangladeshi respectively.> -wrap-foot> *Coefficients were also adjusted for a random effect for practice.
We used mixed linear regression models that included patient socio-demographic variables (categorical age, ethnicity, gender, self-rated health status, self-reported presence of a mental health condition and quintiles of deprivation) as fixed effects and a random effect for practice.
After adjusting for age, gender, deprivation, self-rated health status, self-reported presence of a mental health condition, and a random effect for practice, we found that Bangladeshi respondents on average reported poorer communication than their White British counterparts (-2.9; 95%CI -4.2, -1.6).
Since our endpoint was referral of individual patients, but the unit of randomization was general practice, all models assessing the treatment effect incorporated a random effects component for practice, to take account of this cluster randomization [ 9].
4 10 Rather, our aim has been to take a large representative sample of the UK population being tested in the primary care setting and study the practical effects for clinical practice and for patients of a change from the use of the MDRD formula to the CKD-EPI formula.
These models incorporated the structure of the data by including random effects for person, practice and year.
